In a highly competitive market, olefin polymerization reaction systems are pushed ever closer to their operating constraints in order to increase production in existing equipment. Operating close to the constraints, the possibility of a reactor upset becomes greater. An upset is a reaction which deviates from normal operating constraints to a degree which requires a correction in production rate to prevent a significant economic loss by way of a shutdown, production of unusable product, loss of raw material, or the like. Thus, it is important to control upsets which would otherwise require a shutdown or other drastic or unusual operating action unless an effective amount of polymerization retarding agent is introduced imminently. Typically an upset could involve a significant increase in temperature in the reacting system, which may escalate the reaction further out of control and/or dangerously soften the product particles, causing them to stick together and from large unmanageable agglomerates. Unusual pressure values, superficial gas velocities, or static effects can also be harbingers of upsets, although not as common or always as dependable as temperatures. Upsets can also be the result of influences that are not measured or predicted, such as variations in reactant feed quality, irregular catalyst flow(s), and equipment malfunction. To maintain reactor operation and avoid expensive shutdowns, polymerization retarders can be fed into the reactor during or following an upset once it is detected, but manual intervention to introduce fast acting polymerization retardants may often not be timely enough or accurate enough to avoid the adverse consequences of major process upsets. Human monitoring cannot be expected to suffice where combinations of rapidly changing variables must be interpreted to anticipate runaway conditions.
One method for controlling reactors uses neural net based controllers. A neural network is an interconnected assembly of simple processing elements, units or nodes, whose functionality is loosely based on the animal neuron. The processing ability of the network is stored in the inter-unit connection strengths, or weights, obtained by a process of adaptation to, or learning from, a set of training patterns. In neutral net systems, a collection of process inputs and controllers are simultaneously used and controlled because of the nature of neural nets to achieve the improvement. Thus, when the controller is turned off, there is no effective control of resin properties, control of pressure, control of production rate, optimization of production rate. When the controller is activated, the multiple inputs and process variables are evaluated and the collective nature of these values results in a control effort. It would be desirable to provide a method that can control process variables without using a neural net approach.